This thesis is concerned with the finite sample properties of some of the most widely used two-stage estimators in econometrics. Specifically we examined the use of two-stage estimation in linear regression models with AR(1) errors and in linear regression models with unobservable expectations variables. These are organized into three chapters. In chapter 1, we considered the linear regression model with AR(1) errors and proposed a jack-knifed estimator which reduces the over-rejection problem associated with conventional estimators. In chapter 2, we examined three of the most widely studied expectations models. For each model finite sample properties of two-stage estimators are derived. They served as the basis for various comparative studies of competing two-stage estimators and results useful for the selection of estimators are obtained. In chapter 3, expectations models with AR(1) errors are considered. For this case a two-step GLS estimator is proposed and an exploratory Monte Carlo study confirmed that this new estimator can lead to substantial efficiency gain even in small samples. This thesis also includes an introduction where a useful review of two-step estimation in econometrics is given.